Run an AI-Augmented Ideation Workshop

Workflow area: Ideation & Critique · Tier: Foundation · Hands-on time: ~25 minutes prep

Workshop facilitation has three time sinks — priming the room, clustering the wall, and writing it all up — and none of them are where the value is. The value is the ten minutes when eight humans with different context diverge honestly. AI is superb at the sinks and corrosive at the center: seed the wall with model-generated stickies and the room converges on them, because a written idea reads as a sanctioned idea. This playbook sequences AI around the human moment instead of into it.

When to use — and when not to

• Use it for divergence workshops with 4+ peopleHMW generation, solution sketching, assumption mapping; the clustering and summary gains scale with sticky count.
• Use it when facilitation falls on you and you also need to contributeAI takes the mechanical facilitation load so you can think.
• Use it for distributed and async workshopssorting and summarizing are what make a 40-sticky async board readable across timezones.
• Skip it for decision meetingschoosing between options is judgment and politics; automating it satisfies no one and hides accountability.
• Skip the AI entirely for charged topicsa workshop processing a failed launch or reorg needs human hands on every sticky; summarization that smooths pain flattens the point.

The recipe

1. 1.

   Prep the frame with AI before anyone arrives. Draft the workshop skeleton in Claude, then build it in FigJam:

   Copy

   Design a 60-minute ideation workshop for [N] people on this problem:
   
   Problem: [ONE PARAGRAPH — the user problem, not the solution]
   What we know: [2-3 EVIDENCE BULLETS FROM RESEARCH]
   Output needed: [e.g. "8-10 candidate HMWs, top 3 voted, each with
   first-step owner"]
   
   Produce: agenda with timings, 3 warm-up prompts, facilitation
   script for the divergence block, and the voting method. The agenda
   must put SILENT INDIVIDUAL divergence before any group discussion,
   and all AI-generated content after human divergence is complete.
   

2. 2.

   Run silent human divergence first — AI stays holstered. Individual sticky writing, no talking, no AI on the board. This is the anti-anchoring rule in action: the model's ideas are averages of everyone's products, and averages shown early pull a room to the middle.

3. 3.

   After the wall is full, let AI diverge too — labeled. Now use FigJam AI (or Miro AI on a Miro board) to generate additional HMWs or solution stickies from the problem statement. Give them a distinct color and say so out loud. Two things happen: gaps humans missed get covered, and the room gets to beat the machine — which reliably sharpens human contributions rather than replacing them.

4. 4.

   Cluster with AI, adjudicate as a group. Run FigJam AI's sort-stickies on the full wall, then walk the clusters aloud. Expect to fix them — good adjudication sounds like:

   Copy

   Cluster "Notifications" — actually two ideas: reminding approvers
   vs. removing the need for approval. Split it.
   Cluster "Misc" (6 stickies) — three belong in "Trust", two are
   duplicates, one is a new theme the sort missed: "audit anxiety".
   

   The mis-sorts are informative: a sticky the model can't place is often the original idea in the room.

5. 5.

   Draft the summary before the room empties. Ask FigJam AI to summarize the board, then paste into Claude with the vote results for the send-out note:

   Copy

   Turn this workshop board summary and vote tally into a half-page
   readout: the problem, top 3 voted directions with vote counts, the
   named ideas that got no votes but the facilitator flagged as worth
   revisiting, and next step + owner per direction. Do not add ideas
   that are not on the board.
   [PASTE BOARD SUMMARY + VOTES]
   

   Read it once against the wall before sending — the summary is the workshop's memory, and an invented bullet becomes false history by Friday.

Make it yours

• Miro AI end-to-end: if your org lives in Miro, the same sequence maps directly — its clustering and summarize features slot into steps 4 and 5, and its templates cover step 1's skeleton.
• Claude as facilitation co-pilot: keep a side chat open during the session; when energy dips, ask for a reframing prompt on the spot ("give me three provocations that attack the group's most common assumption on this wall").
• Async-first version: post the problem and evidence Monday, silent divergence runs all week on the board, AI generation + clustering Thursday, 30-minute voting call Friday — the anchoring rule holds even harder async, since early stickies anchor everyone who arrives later.

Quality check

• AI stickies came after human stickies, and are visually distinctif you can't tell which ideas were the machine's, provenance is gone and so is the anchoring control.
• Clusters were adjudicated aloud, not acceptedat least one cluster split, merged or renamed by the room; untouched sorting means nobody looked.
• The summary contains only what's on the wallspot-check every bullet against a real sticky; summarizers pad with plausible ideas nobody had.
• Unclusterable stickies got attentionthe outliers are disproportionately where the original thinking is.
• Zero-vote ideas were preserved, not deletedthis round's noise is next quarter's answer; archive the full board.